fix(streaming): accept bytearray and memoryview audio in stream()

[RaphaelFakhri]

Summary

RealTimeTranscriber.stream() and AsyncRealTimeTranscriber.stream() only recognize bytes as a single buffer. A bytearray or memoryview falls through to the iterable branch and is enqueued one int at a time. The write thread then raises ValueError: Attempted to send invalid message: <class 'int'>. In the sync client the thread dies silently and no audio is sent. In the async client the write task fails with an unretrieved exception.

Changes

• stream() accepts bytes, bytearray and memoryview, and copies the latter two to bytes before enqueuing.
• Iterables (and async iterables) of bytearray or memoryview chunks get the same conversion.
• The conversion lives in a shared _as_bytes_chunk helper in streaming/v3/_base.py.

This matches how transcribe_live() and the dual-channel streamer already treat bytearray and memoryview.

Testing

• Added sync tests (test_streaming.py) and async tests (test_streaming_async.py) for a single buffer and an iterable of chunks, parametrized over bytearray and memoryview.
• Before the change: pytest tests/unit/test_streaming.py tests/unit/test_streaming_async.py -k bytes_like reports 6 failed.
• After the change: 6 passed. pytest tests/unit reports 739 passed.
• ruff format --check assemblyai/streaming is clean.